Technical

Sandbox

Isolated environment for controlled AI testing and evaluation.

How the laws define this

Regulators and policy frameworks conceptualize a sandbox as a controlled, supervised, and time-limited testing environment designed to experiment with, test, and validate artificial intelligence systems prior to full-scale or wider deployment. These environments prioritize safe experimentation, risk containment, and monitored evaluation under regulatory oversight. However, these instruments diverge on whether the framework offers regulatory flexibility or demands strict compliance validation, as well as on sectoral scope. The Hong Kong Monetary Authority Generative AI Sandbox and the startup program Інноваційний Sandbox для стартапів у сфері ШІ та блокчейну offer relaxed regulatory requirements, adapted oversight, or expert consultations to foster innovation. In contrast, the National AI Sandbox (first Spanish EU AI sandbox) focuses explicitly on evaluating regulatory compliance through monitored telemetry, expert reviews, and iterative remedial actions. Furthermore, while the Villani Report and the Hong Kong Monetary Authority Generative AI Sandbox limit testing to sector-specific applications such as banking, instruments like the Malaysia Artificial Intelligence Consortium establish environments for general pre-deployment testing and conformity checks.

Synthesised from the 8 statutory definitions below. Each one is quoted in full, with its source.

How jurisdictions differ

What is distinctive in each definition, binding law in force first.

Jurisdiction	Instrument	Status	What is distinctive
Spain	National AI Sandbox	In Force	Adds regulatory oversight with monitored telemetry, expert review, and iterative remedial actions for high-risk systems under EU AI Act rules.
Chile	Action Plan of the National AI Policy	In Force	Widens scope to inform policy, ensure safety, and enable iterative regulatory learning without full-scale deployment.
Ukraine	Innovation Sandbox for AI and Blockchain Startups in Ukraine	In Force	Limits to experimental environment for AI and blockchain startups providing consultations and expertise to boost competitiveness.
Malaysia	Malaysia Artificial Intelligence Consortium (MAIC)	Adopted	Limits to isolated test environment by consortium for pre-deployment conformity checks and safety issue identification in universities and industry.
Mexico	Toward an AI Strategy in Mexico: Harnessing the AI Revolution	Adopted	Adds trialing of business models and regulatory approaches plus testing impact-assessment procedures as white paper recommendations.
France	Giving Meaning to Artificial Intelligence: For a National and European Strategy	Adopted	Specifies sector-specific use with sensitive or real-world data under tailored governance, safety, and data-access rules.
Lithuania	National AI Strategy update / National AI Governance Forum	Adopted	Adds Innovation Agency operation of a supervised regulatory testing environment for restricted, time-bound testing of AI systems with defined safeguards and pre-agreed mitigation measures.
Hong Kong	Hong Kong Monetary Authority Generative AI Sandbox	Superseded	Limits to a risk-managed environment for financial institutions and technology firms to test GenA.I. solutions in the banking sector with supervisory oversight.

What it triggers

In Spain, participants must deploy, test and validate compliance and mitigation measures under supervised oversight in the National AI Sandbox.source
Under Lithuania's National AI Strategy update, the sandbox describes restricted, time-bound testing with defined safeguards.source

Definitions (8)

A controlled, isolated test environment established by MAIC for safe experimentation and evaluation of AI systems, enabling pre‑deployment testing, conformity checks and identification of safety issues prior to wider deployment.

Malaysia Artificial Intelligence Consortium (MAIC)Definition 1 of 8

A regulated, time-limited testing environment that allows innovators to trial AI systems, business models, or regulatory approaches under supervision and with tailored safeguards. The white paper recommends sandboxes to learn about regulatory impacts and to test impact-assessment procedures before wider deployment.

Hacia una Estrategia de IA en México: Aprovechando la Revolución de la IA (Toward an AI Strategy in Mexico)Definition 2 of 8

A regulated, controlled environment where AI systems can be tested and piloted under monitored conditions and defined risk controls to inform policy, ensure safety, and enable iterative regulatory learning without full-scale deployment.

Plan de Acción para la Política Nacional de Inteligencia Artificial (Action Plan of the National AI Policy)Definition 3 of 8

A sector-specific regulated experimental environment enabling supervised testing and experimentation with AI systems (including use of sensitive or real-world data) under tailored governance, safety and data‑access rules to foster innovation while containing risks.

Villani Report ("Donner du sens à l'intelligence artificielle") - Mission report and recommendations on AI (Cédric Villani), March 2018Definition 4 of 8

A controlled, time-limited and supervised testing environment established by the Decree where participants can deploy, test and validate compliance and mitigation measures under regulatory oversight, including monitored telemetry, expert review and iterative remedial actions.

National AI Sandbox (first Spanish EU AI sandbox / Entorno de pruebas de Inteligencia Artificial)Definition 5 of 8

A controlled, risk-managed environment established by regulators to allow financial institutions and technology firms to test innovative products, services, or business models under relaxed regulatory requirements or with close supervisory oversight. The HKMA's GenA.I. Sandbox specifically facilitates the development and testing of GenA.I. solutions in the banking sector.

Hong Kong Monetary Authority Generative AI SandboxDefinition 6 of 8

A supervised regulatory testing environment operated by the Innovation Agency that offers restricted, time‑bound testing of AI systems with defined safeguards, participant admission rules, monitoring and reporting obligations, and pre‑agreed mitigation measures to enable safe experimentation, including of high‑risk or novel systems.

National AI Strategy update / National AI Governance ForumDefinition 7 of 8

A 'Sandbox' refers to a controlled, experimental environment where innovative products, services, or technologies can be tested and refined under expert guidance, often with adapted regulatory oversight. In this program, it provides consultations and expertise to AI/blockchain teams.

Інноваційний Sandbox для стартапів у сфері ШІ та блокчейнуDefinition 8 of 8

Not to be confused with

Safe space/sandbox

Safe space/sandbox emphasizes experimentation generally, while these definitions add regulatory supervision, time limits or specific sectors.

Sandbox/testbed

Sandbox/testbed covers both concepts as experimental environments, but these specify regulatory oversight or policy-driven controls.

Questions and answers

What is required in Spain's National AI Sandbox?

Participants deploy, test and validate compliance and mitigation measures under regulatory oversight, including monitored telemetry, expert review and iterative remedial actions. It targets high-risk systems under EU AI Act rules.

Does a sandbox impose legal duties?

Only binding instruments like Spain's Decree impose duties such as supervised testing and validation. Policies like Chile's or Ukraine's recommend or describe sandboxes without binding requirements.